Low complexity distributed intelligent cooperative mimo signal processing method and system

Abstract

This invention discloses a low-complexity distributed intelligent cooperative MIMO signal processing method and system, including: using a selected distributed unit to perform preset detection steps to obtain an estimated vector until the number of iterations reaches a preset iteration threshold L; sending the estimated vector after L iterations to a central processing unit, where it is quantized and estimated to obtain the final detection signal; the preset detection steps include: processing the received radio frequency signal to obtain a received signal vector; obtaining a local channel matrix based on a known pilot sequence; calculating a dynamic step size according to an adaptive dynamic step size adjustment strategy; updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration; selecting the next distributed unit using a multi-step conditional random sampling strategy; and sending the updated estimated vector to the selected next distributed unit. This invention reduces computational complexity and data interaction overhead while maintaining detection performance.

Description

Technical Field

[0001] This invention belongs to the field of signal detection algorithm technology for Extra-Large Scale Multiple Input Multiple Output (XL-MIMO) systems in wireless communication, specifically involving a low-complexity distributed intelligent cooperative MIMO signal processing method and system. Background Technology

[0002] With the exponential growth of user numbers and data traffic in wireless communication systems, MIMO technology has become one of the core technologies for fifth-generation (5G) and even sixth-generation (6G) mobile communication systems due to its outstanding advantages in improving spectral efficiency, energy efficiency, and transmission reliability. In massive MIMO systems, base stations are equipped with hundreds or thousands of antennas to serve multiple users simultaneously, achieving unprecedented spatial diversity gain. However, as the antenna scale continues to expand, traditional centralized baseband processing architectures face severe challenges: baseband data from all antennas needs to be converged to a central processing unit for detection and demodulation, which leads to a sharp increase in computational complexity and a surge in the bandwidth requirements for data interconnection from antenna modules to the central processing node, forming a system bottleneck. For example, performing zero-forcing (ZF) detection or minimum mean square error (MMSE) detection with hundreds of antennas requires inverting a large-dimensional matrix, which is highly complex and difficult to implement in real time.

[0003] To alleviate the aforementioned problems, various distributed signal detection algorithms have been proposed, such as Decentralized Baseband Processing (DBP) architecture, Partially Decentralized (PD) architecture, and Fully Distributed (FD) architecture. Under a distributed architecture, various communication topologies can be used to achieve connection and cooperation between distributed units (DUs), including daisy-chain, ring, and star topologies. For example, in a daisy-chain structure, each DU transmits and processes signal estimation unidirectionally in a chain-like sequence, forming a pipeline-like detection mechanism with low interconnect bandwidth requirements and good scalability. In a ring structure, multiple DUs are connected end-to-end, and the signal is updated cyclically within the ring. In a star structure, each DU is directly connected to a central fusion node, which collects information from each DU and fuses it to obtain the detection result. While these architectures offer scalability, they also bring new challenges, and existing distributed uplink detection schemes each have their limitations. For example, ADMM (Alternating Direction Method of Multipliers) and CG (Conjugate Gradient) methods under the DBP architecture typically require frequent iterative information exchange, resulting in high communication complexity. Detection schemes under the PD and FD architectures suffer from performance degradation and insufficient scalability. Furthermore, in daisy-chain or other topology-based recursive algorithms, methods such as recursive least squares, stochastic gradient descent, and decentralized Newton can update estimates per DU, but these algorithms often have specific requirements on the number of antennas within each DU and suffer from poor convergence performance.

[0004] In the field of distributed large-scale MIMO detection algorithms, the Kaczmarz iterative method has attracted attention due to its simplicity and row-by-row update characteristics. The classic Kaczmarz algorithm updates the detection vector by projecting each row of the channel matrix sequentially, which can be regarded as an iterative method for solving a system of linear equations. However, for inconsistent linear systems with noise interference, the traditional Kaczmarz algorithm suffers from an unavoidable steady-state error, which limits the accuracy of convergence to the true solution and leads to a decrease in detection performance. To improve convergence, researchers have proposed the Randomized Kaczmarz (RK) algorithm, which updates by randomly selecting a row (or an antenna) each time, and statistically can achieve exponential convergence in the desired sense. However, the RK algorithm converges to a non-zero bias in the presence of noise, making it difficult to obtain an unbiased estimate. Similarly, block-based expansion algorithms such as Randomized Block Kaczmarz (RBK) randomly select a set of antennas (a sub-block of the matrix) for updating in each iteration, which can accelerate convergence. However, there is a problem that the same subset of antennas may be repeatedly selected in different iterations, leading to unbalanced data utilization and affecting the convergence speed and performance. To address this, the Multi-step Conditional Randomized Block Kaczmarz (MCRBK) algorithm has been proposed in the literature. MCRBK introduces a "condition" mechanism into the iterative sampling strategy: ensuring that the data block corresponding to each DU is selected once within a period consisting of several iterations, thereby avoiding repeated sampling of the same data block. By selecting an appropriate iteration step size, this method overcomes the convergence problem of the traditional Kaczmarz method, achieves exponential convergence for linear detection solutions, and has global convergence without additional conditional constraints. The MCRBK algorithm is flexibly applicable to various distributed topologies such as rings and stars, and has good scalability. However, MCRBK requires calculating the pseudo-inverse of the selected submatrix in each iteration, resulting in high computational complexity: the computational complexity of a single iteration is approximately... (in The number of antennas per DU (Total number of user antennas). In ultra-large-scale scenarios such as XL-MIMO, even though MCRBK has good performance, there is still an urgent need to further reduce its complexity overhead.

[0005] In summary, there is an urgent need for a distributed XL-MIMO signal detection method that can achieve high detection accuracy with low computational complexity and low data transmission bandwidth. This method should avoid the limitations of existing solutions, such as high communication overhead, stringent requirements for DU configuration, or convergence errors, while also possessing general adaptability to different topologies and real-world scenarios. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a low-complexity distributed intelligent cooperative MIMO signal processing method and system, which significantly reduces computational complexity and data interaction overhead while maintaining detection performance, and can be applied to distributed cooperative communication scenarios in future 6G networks.

[0007] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0008] In a first aspect, the present invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method, comprising:

[0009] The selected distributed unit performs a preset detection step to obtain the estimated vector until the number of iterations reaches a preset iteration threshold. Each distributed unit is accessed once within a sampling period;

[0010] Will pass The estimated vector of the next iteration The data is sent to the central processing unit (CPU) for quantization and vector estimation. The final detection signal is obtained. ;

[0011] The preset detection steps include:

[0012] The received radio frequency signal is processed to obtain the received signal vector;

[0013] Local channel estimation is performed using known pilot sequences to obtain the local channel matrix;

[0014] The dynamic step size is calculated based on the predetermined adaptive dynamic step size adjustment strategy.

[0015] The estimated vector is updated using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration.

[0016] The next distributed unit is selected using a multi-step conditional random sampling strategy.

[0017] The updated estimated vector is sent to the next selected distributed unit.

[0018] In conjunction with the first aspect, optionally, the step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes:

[0019] In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows:

[0020] ,

[0021] In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

[0022] In conjunction with the first aspect, optionally, the formula for calculating the dynamic step size is:

[0023] ,

[0024] In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

[0025] In conjunction with the first aspect, optionally, the method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration employs the following calculation formula:

[0026] ,

[0027] In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Represents the local channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

[0028] In conjunction with the first aspect, optionally, the final detection signal The calculation formula is:

[0029] ,

[0030] In the formula, Round to the nearest whole number constellation points, for A set consisting of constellation points corresponding to each transmitted signal.

[0031] Secondly, the present invention provides a low-complexity distributed intelligent cooperative MIMO signal processing system, comprising several distributed units and a central processing unit; each distributed unit has a bidirectional interactive link with the central processing unit; each distributed unit includes a receiving antenna array, a radio frequency processing unit, a channel estimation unit, and a signal detection unit;

[0032] Each distributed unit executes a preset detection step until the preset number of iterations reaches a preset threshold. ;;

[0033] The central processing unit receives the data. The estimated vector of the next iteration and for the estimated vector Quantization processing is performed to obtain the final detection signal. ;

[0034] The preset detection steps include:

[0035] The receiving antenna array is used to receive radio frequency signals;

[0036] The radio frequency processing unit processes the received radio frequency signal to obtain the received signal vector;

[0037] The channel estimation unit uses a known pilot sequence to perform local channel estimation and obtain a local channel matrix.

[0038] The signal detection unit calculates the dynamic step size according to a predetermined adaptive dynamic step size adjustment strategy; updates the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration; selects the next distributed unit using a multi-step conditional random sampling strategy; and sends the updated estimated vector to the selected next distributed unit.

[0039] In conjunction with the second aspect, optionally, the step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes:

[0040] In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows:

[0041] ,

[0042] In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

[0043] In conjunction with the second aspect, optionally, the formula for calculating the dynamic step size is:

[0044] ,

[0045] In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

[0046] In conjunction with the second aspect, optionally, the method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration employs the following calculation formula:

[0047] ,

[0048] In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Represents the local channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

[0049] In conjunction with the second aspect, optionally, the final detection signal The calculation formula is:

[0050] ,

[0051] In the formula, Round to the nearest whole number constellation points, for A set consisting of constellation points corresponding to each transmitted signal.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] This invention proposes a low-complexity distributed intelligent cooperative MIMO signal processing method (EDRID method) and system for uplink signal detection in XL-MIMO systems. While maintaining near-linear optimal detection performance, this invention significantly reduces algorithm complexity and communication overhead, and offers good flexibility and scalability. Compared with existing distributed detection schemes, this method not only achieves higher detection accuracy and lower processing complexity, but also requires less inter-unit data transmission bandwidth. Furthermore, theoretical analysis proves its global convergence characteristics, making it adaptable to various types of distributed MIMO system applications.

[0054] Specifically, compared to existing distributed detection algorithms, this invention has the following innovations and benefits: First, the EDRID method removes the matrix pseudo-inverse calculation during the iteration process and directly uses the conjugate transpose of the channel submatrix for updating, reducing the computational complexity of each iteration from the original... The order of magnitude decreased to This significant reduction in complexity effectively alleviates the computational bottleneck of distributed detection. Secondly, this invention introduces an adaptive dynamic step-size adjustment strategy to overcome the convergence bias problem inherent in algorithms with fixed step sizes. By adjusting the step size according to the iteration process, the EDRID method can gradually eliminate steady-state errors while ensuring rapid convergence. Theoretically, it has been proven that after sufficient iterations, the algorithm can approximate and precisely converge to the least squares (LS) solution. This means that the EDRID method achieves low complexity while maintaining detection performance consistent with ideal linear detection results. Thirdly, this invention incorporates a multi-step conditional sampling mechanism (i.e., a multi-step conditional random sampling strategy) during the random iteration process: the data from each distributed unit is used sequentially in several iterations of each round, avoiding information redundancy caused by selecting the same unit in consecutive iterations. By considering the sampling results of multiple previous iterations, the random iterative update of the EDRID method gradually exhibits determinism, enabling stable and rapid convergence under different decentralized topologies such as rings, stars, and daisies. In summary, with the aforementioned improvements, the EDRID method, with its low complexity, strong adaptability, and good scalability, has become a powerful solution for addressing the distributed detection problem of XL-MIMO in the 6G era. The method of this invention achieves an ideal balance between performance, complexity, and data bandwidth, and is widely applicable to distributed large-scale MIMO scenarios such as XL-MIMO and Cell-Free MIMO. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0056] Figure 1 This is a schematic diagram of a classic uplink MIMO system scenario.

[0057] Figure 2 This is a ring-distributed system architecture diagram of a low-complexity distributed intelligent cooperative MIMO signal processing method according to the present invention.

[0058] Figure 3 This is a star-shaped distributed system architecture diagram of a low-complexity distributed intelligent cooperative MIMO signal processing method according to the present invention.

[0059] Figure 4 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 256×32 XL-MIMO system.

[0060] Figure 5 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 256×64 XL-MIMO system.

[0061] Figure 6 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 512×64 XL-MIMO system.

[0062] Figure 7 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 512×128 XL-MIMO system.

[0063] Figure 8 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 1024×256 XL-MIMO system.

[0064] Figure 9 This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method. A comparison chart of the bit error rate trends of various detection schemes in a 16-QAM 512 × 256 XL-MIMO system;

[0065] Figure 10 This is a flowchart illustrating a low-complexity distributed intelligent cooperative MIMO signal processing method according to an embodiment of the present invention.

[0066] Figure 11 This is a flowchart illustrating the preset detection steps of one embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0068] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0069] Example 1

[0070] This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing method, also known as the Efficient Distributed Randomized Iterative Detection (EDRID) method. Figure 10 As shown, it includes the following steps:

[0071] (1) Use the selected distributed unit to perform the preset detection steps to obtain the estimated vector until the number of iterations reaches the preset iteration threshold. Each distributed unit is accessed once within a sampling period;

[0072] (2) After The estimated vector of the next iteration The vector is sent to the central processing unit (CPU), which then quantizes the estimated vector. The final detection signal is obtained. ;

[0073] Among them, such as Figure 11 As shown, the preset detection steps include:

[0074] The received radio frequency signal is processed to obtain the received signal vector;

[0075] Local channel estimation is performed using known pilot sequences to obtain the local channel matrix;

[0076] The dynamic step size is calculated based on the predetermined adaptive dynamic step size adjustment strategy.

[0077] The estimated vector is updated using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration (i.e., the estimated vector output by the previously selected distributed unit).

[0078] The next distributed unit is selected using a multi-step conditional random sampling strategy.

[0079] The updated estimated vector is sent to the next selected distributed unit.

[0080] In one specific embodiment of the present invention, the step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes:

[0081] In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula of the sampling result is used to select the next distributed unit (i.e., to select the sampling result of the sampling result). (a distributed unit), the calculation formula is:

[0082] ,

[0083] In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

[0084] In one specific embodiment of the present invention, the formula for calculating the dynamic step size is:

[0085] ,

[0086] In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

[0087] In one specific embodiment of the present invention, the method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration is calculated using the following formula:

[0088] ,

[0089] In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Represents the local channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

[0090] In one specific embodiment of the present invention, the final detection signal The calculation formula is:

[0091] ,

[0092] In the formula, Round to the nearest whole number constellation points, for The set of constellation points corresponding to each transmitted signal (i.e., the constellation point set).

[0093] The low-complexity distributed intelligent cooperative MIMO signal processing method in this embodiment of the invention is mainly applied to uplink signal detection scenarios in large-scale MIMO systems. For example... Figure 1 As shown, in a massively multi-channel MIMO system, including root transmitting antenna and For a given receiving antenna, K ≤ N, the received signal vector is... and the transmitted signal vector transmitted by the transmitting antenna The relationship between them is shown in the following formula:

[0094] ,

[0095] In the formula, Represents the local channel matrix, with dimension . ; This represents Gaussian white noise interference, with dimensions of . , follows the mean The variance is The complex Gaussian distribution; This represents the transmitted signal vector sent by the user side, with dimension 1. Each element belongs to a set of constellation points modulated by M-QAM. . This represents the received signal vector at the base station, with dimension . To recover the transmitted signal vector in such an XL-MIMO system. Maximum Likelihood (ML) detection aims to solve the following Integer Least Squares (ILS) problem:

[0096] ,

[0097] In the formula, This represents the detection signal obtained through ML detection. This represents the square of the 2-norm. This represents the minimum optimization operator.

[0098] Such problems are nondeterministic polynomial problems, theoretically difficult to solve. Existing linear signal detection schemes are based on... In this case, the problem is transformed into solving the following least squares (LS) problem:

[0099] ,

[0100] In the formula, This represents the detection signal obtained through a linear signal detection scheme;

[0101] Traditional linear signal detection schemes can approximate the solution of ML detection very well, and the estimated signal is:

[0102] and ,

[0103] In the formula, and These represent the estimated signals obtained through linear detection by ZF and MMSE, respectively;

[0104] Finally, based on the constellation point set Quantification and The final judgment was obtained and :

[0105] and ,

[0106] In the formula, and These represent the final detection signals obtained after linear detection and decision-making by ZF and MMSE, respectively. This represents a distributed unit.

[0107] However, the matrix inversion operation involved in this linear solution has a computational complexity of O(n log n). This approach is unusable in XL-MIMO systems. Furthermore, as the antenna and user scale in XL-MIMO systems continue to increase, a decentralized baseband processing architecture capable of distributing computational load is urgently needed in the uplink to perform signal detection tasks. In other words, computational tasks should be divided into multiple geographically dispersed or logically separated units to be completed in parallel, reducing the processing pressure at a single point and avoiding the bandwidth required for centrally aggregating massive amounts of data.

[0108] In this invention, the base station side uses a ring-star topology or other distributed frameworks for connection, such as... Figure 2 and Figure 3 As shown. Base station side. The root receiving antenna is evenly divided into In each distributed unit, these distributed units are recorded as an index set. , Indicates the first Each distributed unit (DU) contains [number] distributed units. One antenna and one independent hardware computing unit, satisfying During XL-MIMO system operation, all DUs perform local channel processing in parallel, and then execute the detection algorithm sequentially according to the topology connection. Within each DU, its... The radio frequency signal received by the antenna is amplified, filtered, down-converted, and analog-to-digital converted by the radio frequency processing unit to obtain a digital baseband signal and a received signal vector. Then, the channel estimation unit performs local channel estimation using the pilot signal (i.e., performs local channel estimation using a known pilot sequence), obtaining a dimension of... Local channel matrix And send it to the signal detection unit. In the first In the next iteration, the signal detection unit utilizes the local channel matrix. And the received signal vector (i.e., the baseband signal). Combined with the signal estimation results from the previous distributed unit (i.e., the estimated vector obtained after the previous iteration), signal detection is performed based on the EDRID method proposed in this invention. Finally, the first... In each iteration, the final estimated vector within the distributed unit is passed to the central processing unit, where the channel decoding task is completed through corresponding quantization operations. Specifically, for example in a ring / chain structure, the first distributed unit first determines the channel decoding task based on the initial estimated vector. and the received signal vector observed by itself Local channel matrix The updated estimated vector is obtained And then pass it to the second distributed unit, which then uses the received signal vector Local channel matrix Estimated vector Update to estimated vector The result is passed to the next unit, and so on in a loop. In a star topology, the central node coordinates and schedules each DU to update the estimated vector sequentially. Each DU performs the update and sends the result back to the central node, which then distributes it to the next DU. After several iterations, the final estimated vector is passed to the central processor, where the channel decoding task is completed through the corresponding quantization operation. This architecture has good scalability: increasing the number of DUs linearly distributes the computational load, and the serial link transmission only involves the estimated vector. The bandwidth required for round-by-round transmission increases linearly with the number of DUs. Throughout the process, unlike centralized schemes, it is not necessary to collect all raw antenna data or full Channel State Information (CSI) to a central location, which greatly reduces the amount of interconnection communication.

[0109] Within the aforementioned distributed processing framework, the EDRID method proposed in this invention is used to perform iterative signal detection across DUs. For ease of explanation, we first consider the application of the traditional Kaczmarz method in distributed detection, which only involves row calculations of the channel matrix. Computation in distributed units:

[0110]

[0111] Each DU contains only one single antenna (i.e. In this case, the local channel matrix of each DU can be represented as... Under ideal, noise-free conditions, the traditional Kaczmarz method can converge to an exact solution for uniform linear systems of equations, but it fails for incompatible linear systems corrupted by noise. In this context, the convergence of the traditional Kaczmarz method cannot be guaranteed, resulting in poor performance when directly applied to distributed noisy MIMO detection.

[0112] To improve convergence accuracy, the Randomized Kaczmarz (RK) algorithm, which incorporates random sampling, is introduced. In the RK algorithm, a single DU (Data Unit) is randomly selected for each iteration for the aforementioned update. Theoretical analysis shows that, in the absence of noise, the expected convergence rate of the RK algorithm is exponential. However, the presence of noise introduces a systematic bias into its convergence. A further improvement is to consider the Randomized Block Kaczmarz (RBK) algorithm, which uses data from multiple antennas simultaneously. In each iteration, a single DU containing multiple antennas is randomly selected. The DU of the root antenna, using the observation of this DU To update the current estimated vector. In practical applications, the channel matrix... The row indicators are divided into different subsets. Then, in each iteration, from the index set... Randomly select a subset , Represented by a uniform distribution:

[0113] ,

[0114] In the formula, express The selection follows a uniform distribution;

[0115] Obtain the corresponding row matrix Then project the current result onto The iteration is performed on the solution space, that is:

[0116] ,

[0117] The RBK algorithm uses block information to accelerate convergence, but due to the random selection of DUs, adjacent iterations may choose the same DU. This situation leads to some DU data being reused while other DUs remain unused for a long time, which is not conducive to global equilibrium convergence.

[0118] To address this issue, the Multi-Step Conditional Randomized Block Kaczmarz (MCRBK) algorithm was proposed. This algorithm avoids the problem of repeated sampling by adding conditional constraints during the sampling process. The MCRBK algorithm specifies a condition of length... Within the iteration window, each DU's data block is used once; thus, each iteration... Next iteration, all Each element is utilized once. This is equivalent to sequentially connecting DUs in a loop, thus combining the recursiveness of the Kaczmarz method with the pipeline transmission capabilities of a distributed architecture. The iterative update form of the MCRBK algorithm is:

[0119] ,

[0120] in, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix The false rebellion, This refers to the step size chosen for each iteration. Appropriate selection is crucial. It can be proven that MCRBK achieves exponential convergence for linear detection solutions and possesses global convergence for various initial conditions. However, each update of MCRBK still requires computation. pseudo-reversal For dimensions Local channel matrix This step requires The total complexity of the operation, including the multiplication update, is approximately In XL-MIMO scenarios with a large number of antennas and users, this complexity still cannot be ignored.

[0121] The EDRID method of this invention addresses the aforementioned problems. While retaining the multi-step conditional sampling strategy from MCRBK to ensure full utilization of each DU's data, the EDRID method avoids matrix pseudo-inverse calculations in the iterative update formula, significantly reducing algorithm complexity. Specifically, the EDRID method uses the conjugate transpose of each DU's channel matrix... Replaced pseudo-reverse To perform iterative updates, that is:

[0122] ,

[0123] Compared to the MCRBK algorithm, the EDRID method in this invention avoids the cumbersome computation of matrix inversion, reducing the update complexity to the cost of matrix multiplication. Each update multiplication mainly involves operations with a complexity of O(n log n). of The complexity is of Approximately This is the second multiplication. Compared to the MCRBK algorithm... The complexity advantage of the EDRID method proposed in this invention is particularly evident in large-scale systems.

[0124] In addition, a fixed step size is directly adopted. The iterative process of the EDRID method tends to converge to a biased range around the target solution, determined by the step size, rather than the exact solution. To overcome this convergence bias, this invention further introduces a dynamic step size adjustment mechanism into the EDRID method. Dynamic step size The update magnitude gradually decreases with each iteration, thus approaching zero and ensuring that the iteration theoretically converges to the least squares solution. Finally, the EDRID method adopts the multi-step conditional sampling of the MCRBK algorithm to maximize the advantages of the distributed architecture. In practice, a multi-step sampling period is typically selected. This ensures that each iteration fully traverses all... Each DU. This means that the EDRID method passes through each DU. After each update, the observation information of all DUs is effectively utilized, and then the next sampling cycle begins. As the number of iterations increases, the EDRID method's dependence on random sampling gradually decreases, and the algorithm's behavior becomes more deterministic, thus accelerating convergence. This mechanism also allows the EDRID method to be easily deployed in various topologies, such as cyclically sampling DUs in a fixed order in a chain or ring structure, or scheduling DUs sequentially by the center in a star structure, all of which meet the requirements of a multi-step conditional sampling strategy. Therefore, the method of this invention has strong topological universality and scalability. Theoretical analysis and simulation results both show that the EDRID method achieves near-optimal detection performance in various XL-MIMO scenarios and significantly reduces computational complexity and interconnect communication burden. For example... Figure 3 As shown, in an XL-MIMO system with a 256×64 antenna configuration, the EDRID method, combined with a dynamic step size, can achieve a bit error rate performance close to that of centralized MMSE detection with an extremely low number of iterations. In contrast, traditional distributed detection algorithms (such as ADMM, CG, DN, etc.) are either too complex to implement or lose convergence or have poor performance in XL-MIMO scenarios. The EDRID method proposed in this invention provides a practical, efficient, and reliable solution for ultra-large-scale distributed MIMO signal detection by cleverly balancing computational efficiency and detection accuracy.

[0125] The following uses a transmitting antenna Receiving antenna Number of distributed units Taking a distributed XL-MIMO system as an example, the specific implementation of the present invention is described.

[0126] Step 1: Distributed XL-MIMO system initialization and channel acquisition. For the transmit antenna... Receiving antenna Distributed XL-MIMO system, Each distributed unit is assigned to Root receiving antenna, the first Local channel matrix of each distributed unit and local received signal vector , , can be represented as:

[0127] ,

[0128] in, The dimension is , The dimension is . Indicates the first root transmitting antenna and the first Channel response between root receiving antennas Indicates the first The received signal on the root receiving antenna.

[0129] Step 2: EDRID Method Initialization. Before starting iterative detection, the algorithm parameters, including the estimation vector, are uniformly initialized in each distributed unit. Dynamic step size And set the threshold for the number of iterations of the EDRID method. The specific operating steps are as follows:

[0130] Step 2.1 Estimate the user signal vector The initial value is set as a zero vector with dimension . .

[0131]

[0132] Step 2.2 sets the initial value of the dynamic step size parameter in the EDRID method as follows:

[0133] ,

[0134] Step 2.3 Set the threshold for the number of iterations This controls the iteration flow of the EDRID method; it can be set here. .

[0135] Step 3: Distributed Random Iterative Detection Process. Within the EDRID framework, each DU (Distributed Random Detection Unit) takes turns iteratively updating the user signal estimation vector in a specific order. The specific operations are as follows:

[0136] Step 3.1 Multi-step conditional sampling sequence initialization. Determine the access order of each distributed unit in a complete iteration loop. Employ a multi-step conditional random sampling strategy to generate the initial DU access order, and initialize the current global estimation vector. Send to the first selected DU.

[0137] Step 3.2 Select the next distributed unit for detection and update the estimate, and continue this process until all units have completed one update before starting a new loop. In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows:

[0138]

[0139] Here, in the formula, This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in each iteration is used to ensure that each DU is accessed once within a sampling period, thereby avoiding the reuse of the same DU's data in consecutive iterations.

[0140] Step 3.3 Iterative update calculation using the EDRID method. Within each selected distributed unit, the user signal vector is estimated using the EDRID method of this invention. and dynamic step size Iterative updates are performed, among which, The current iteration number is calculated using the following steps:

[0141] Step 3.3.1: After each iteration, adjust the step size according to the predetermined dynamic step size strategy. Update the dynamic step size value after each iteration. The methods that can be adopted are:

[0142] ,

[0143] Step 3.3.2: Based on the estimated vector obtained from the previous iteration and the updated dynamic step size Perform random iterative detection and update calculations on the estimated vector:

[0144] ,

[0145] In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

[0146] Step 3.4 Comparison and The size, if Proceed to step 3.2. If If so, proceed to step 3.5.

[0147] Step 3.5 At this point, the process has been completed. The estimated vector of the next iteration It is then passed to the central processing unit and step 4 is executed.

[0148] Step 4: Central Processing Unit Quantizes Estimated Vector By using constellations Rounding to estimate vectors ,get To be used as the final output:

[0149]

[0150] in, Round to the nearest whole number The constellation points.

[0151] Regarding the data transmission bandwidth of this invention:

[0152] Since the EDRID method in this invention is a distributed algorithm, the bandwidth on the interconnecting links needs to be considered. Data transmission bandwidth is determined by the average complex value transmitted on each link, which reflects the actual overhead of the hardware interface. In this invention, because only the updated estimated vector is available... It needs to be transmitted unidirectionally through distributed units, and its data bandwidth requirement is only... .

[0153] When various algorithms such as ADMM (Alternating Direction Method of Multipliers), CG (Conjugate Gradient), MMSE (Minimum Mean Square Error), CD (coordinate descent), DN (decentralized Newton), RLS (Recursive Least Square), SGD (Stochastic Gradient Descent), ASGD (Averaged Stochastic Gradient Descent), SDK (standard distributed Kaczmarz), and MCRBK (Multi-Step Conditional Randomized Block Kaczmarz) are used to replace the EDRID method in the embodiments of this invention, these algorithms either have very high computational complexity or data bandwidth costs, or limit the number of distributed element antennas, making them inflexible to implement. This invention achieves accurate convergence while avoiding matrix pseudo-inverse calculations and maintaining an interconnect bandwidth of only [amount missing]. 3D symbol vector transmission is more suitable for ultra-large-scale decentralized hardware architectures. Considering a comprehensive comparison of detection performance, computational complexity, and transmission bandwidth, the proposed scheme exhibits the best overall performance across various XL-MIMO communication scenarios, as shown in the attached figures. Figures 4-9 This can be proven at the same time. Figures 4-9 The vertical axis in the graph represents the bit error rate (BER), and the horizontal axis represents the ratio of bit energy (Eb) to noise power spectral density (N0). To visually demonstrate the superiority of this scheme, the algorithm of this invention is compared with other existing schemes in terms of complexity, bandwidth, antenna quantity, global convergence, and distributed architecture, as shown in the table below:

[0154]

[0155] In the table: Computational Complexity represents complexity; Data Bandwidth represents bandwidth; Flexible number of antennas in each DU indicates whether the number of receiving antennas in each distributed unit can be flexibly set; Global Convergence represents global convergence; Decentralized Architecture represents distributed architecture; Ring represents ring architecture; Star represents star architecture; Daisy-chain represents daisy-chain architecture. The number of iterations is indicated by FD, PD, and DBP. FD represents fully distributed, PD represents partially distributed, and DBP represents a distributed (or decentralized) baseband processing architecture.

[0156] Existing patents and solutions in distributed XL-MIMO uplink detection typically employ direct matrix inversion or pseudo-inversion to solve the linear detection problem. This approach suffers from high computational complexity and limited convergence accuracy in noisy, inconsistent linear systems. For example, the Multistep Conditional Random Block Kaczmarz algorithm (MCRBK), as an existing solution, requires calculating the pseudo-inversion of the selected channel submatrix in each iteration, resulting in a computational cost of approximately [missing information - likely related to computational complexity]. In XL-MIMO scenarios with extremely large antenna scales, such high complexity is difficult to meet real-time processing requirements. Furthermore, in noisy situations, the fixed-step Kaczmarz iteration converges to a steady-state point deviating from the true solution, resulting in an unavoidable convergence error and decreased detection accuracy. The EDRID method proposed in this invention reduces complexity while eliminating convergence bias, achieving high-precision approximation of the least-squares solution.

[0157] In each iteration of the EDRID method, the pseudo-inverse calculation of the matrix is ​​removed, and the conjugate transpose of the channel submatrix is ​​used directly for updating, thereby significantly reducing computational complexity. Simultaneously, EDRID introduces a dynamic step-size strategy, which gradually eliminates steady-state errors while ensuring rapid convergence of the iterations, achieving accurate convergence of the least squares (LS) solution. The theoretical convergence and effectiveness of EDRID are explained below from four aspects: the upper bound of the convergence error, the step-size selection condition, the error recursion relationship, and the condition for the error to converge to zero.

[0158] Derivation of the upper bound formula for convergence error: The iterative process of the EDRID method can be represented as a stochastic iteration of solving a system of linear equations. For the case containing noise, if the least squares solution is denoted as... Then after The upper bound of the mean square expectation of the error after n random iterations satisfies:

[0159] ,

[0160] in The squared norm of a vector is denoted by the L2 norm. For the first The estimated vector obtained from the second iteration. The iteration step size, It is the convergence factor. This reflects the impact of noise on steady-state error. Let be the initial estimation vector. The above inequality is derived by establishing a recursive relationship for the iteration error and taking the expectation of the noise term: First, the current error is expressed as the previous iteration error minus the projection correction term along the selected row (or sub-block) direction. Then, the squared error norm is expanded and its upper bound is estimated, finally yielding the result as the number of iterations increases. The expression for the increasing and converging upper bound of the error. As shown in the second term of the equation, when... The expectation of the time error norm will approach This is the upper bound of the steady-state error of the algorithm under a fixed step size. In other words, although the error gradually decreases with iteration, it will converge to a deviation region within a certain radius from the least squares solution when noise is present. Under ideal consistency conditions without noise, When the value approaches 0, the above formula degenerates into... This means that the error will be... This is the solution where the scaling factor converges to zero error.

[0161] Derivation of the step size selection range condition: To ensure iteration convergence and minimize steady-state deviation, the iteration step size... Certain range conditions need to be met. Based on convergence analysis, it can be deduced that the range of step size values ​​must satisfy the following inequality:

[0162] ,

[0163] in, For all local channel submatrices Gram matrix The upper bound of the largest eigenvalue, i.e. ( (This refers to the number of distributed units). The above condition is derived from the convergence factor in the formula for the upper bound of the convergence error. The expression that leads to: if we take satisfy Then the convergence factor It will be strictly less than 1, where, For the overall channel matrix The minimum singular value), thus ensuring It decays exponentially. Simply put, the iteration step size... The step size must be less than twice the reciprocal of the largest eigenvalue of each submatrix. This conclusion is consistent with the convergence condition of the classic Landweber iterative method, ensuring that the error correction in each iteration is not excessive and thus avoids divergence. In practical applications, the iteration step size... You can choose an experienced recommended value (e.g.) , (For user dimensions) to achieve a faster convergence speed.

[0164] Error recursion formula and explanation: To gain a deeper understanding of the formation of the upper bound of error, we examine the error recursion relationship of the EDRID iteration. Let the first... The estimated vector for the next iteration is Then the first The estimated vector obtained after the next iteration update for:

[0165] ,

[0166] In the formula, Indicates the randomly selected first Local channel matrix of each DU For the first The received signal vector of each DU. Subtract both sides of the above iterative update equation. The update formula for the error vector can be obtained as follows:

[0167] ,

[0168] This indicates that the new error vector is formed by adding a correction term to the old error vector. The correction term... Using the first The conjugate transpose of the local channel matrix of each DU will be used to determine the current residual. Projecting the vector onto the solution vector space partially eliminates the error. When noise exists in the MIMO system, the received vector can be represented as... ( If the noise vector is denoted by , then the residual contains a noise term, meaning that a single iteration cannot completely eliminate the error. In this case, the above error recursion can be expanded to obtain the change in the error norm:

[0169] ,

[0170] And use equations in the derivation Take the expectation of the noise term and use inequalities. (in Using techniques such as treating the vector as any non-zero vector, an upper bound is applied, ultimately yielding the upper bound form of the convergence error shown in the aforementioned derivation formula. It can be seen that the noise term is... This manifests as residual deviation terms, while This determines the convergence rate of the noise-free portion.

[0171] Analysis of the conditions for error convergence to 0 after multiple iterations: To address the steady-state error problem in fixed step-size schemes, the EDRID method introduces an adaptive dynamic step size. This allows the algorithm to converge the error to zero after multiple iterations. The core idea of ​​the dynamic step-size strategy is to adaptively adjust the step size in each iteration based on the current error state, so that the convergence coefficient and the error constant term decay synchronously. Specifically, this involves setting the two parts of the aforementioned error upper bound formula... and The values ​​are merged and uniformly decreased during the iteration process. Therefore, a relationship with the number of iterations is introduced. Related dynamic step size And select the optimal step size so that the mean square error after each iteration is relative to the dynamic step size. Reaching the minimum. The derivation shows that, when the recursive relation is in effect, the dynamic step size... When the derivative is zero, the optimal step size selection in closed form can be obtained as follows:

[0172] ,

[0173] in, These are constants related to the minimum singular value and the number of elements in the channel matrix. The recursive sequence of the coefficients of the error term is defined as follows: Its initial value is Using the dynamic step size formula described above, the EDRID method achieves its optimal step size in the first step. The update form of the next iteration can be expressed as:

[0174] ,

[0175] And always meet the previous step size range limit. Under the influence of dynamic step size, it can be proven that the error of the EDRID method proposed in this invention will strictly monotonically decrease and eventually eliminate the constant deviation term, achieving accurate convergence of the LS solution. Formally, after... After the next iteration, we have:

[0176]

[0177] Because each item is dynamically adjusted All meet (Global convergence condition) When the number of iterations... When the product term on the right-hand side of the above approaches zero, therefore This indicates that the algorithm of this invention, by gradually reducing the step size, can make the estimated vector converge to the least squares solution with probability 1 after sufficient iterations. Theoretically, EDRID has been proven to converge precisely to the optimal solution of linear detection at an exponential rate, completely eliminating the steady-state error of the fixed step size scheme.

[0178] Compared to the original MCRBK scheme, the EDRID method demonstrates significant advantages in both computational complexity and error control mechanisms:

[0179] In terms of complexity: the original MCRBK algorithm requires calculating the pseudo-inverse of the channel submatrix in each iteration, which involves expensive operations such as matrix multiplication and inversion. For example, for a dimension of... Performing QR decomposition or Gram matrix inversion on a submatrix typically has a time complexity of O(n log n). The order of magnitude is even higher. In contrast, the EDRID method in this invention, because it directly uses the conjugate transpose for iterative updates, only requires matrix and vector multiplication and addition in each iteration, reducing the complexity to linear level. Floating-point arithmetic. Empirical evidence shows that the computational cost of a single EDRID iteration is... The complexity of EDRID is far less than that of MCRBK, which involves pseudo-inverse calculations. Therefore, for problems of the same size, EDRID takes less time per iteration and can complete real-time detection within a finite number of iterations. This complexity advantage is particularly prominent in systems with a huge number of antennas, such as XL-MIMO, which can significantly alleviate the computational bottleneck of distributed detection.

[0180] Regarding error convergence and control: Although the MCRBK algorithm accelerates convergence through multi-step conditional sampling, theoretical analysis shows that its random iteration will eventually converge to a region at a certain distance from the LS solution, meaning that there is a non-negligible lower bound for error under a fixed step size. This implies that for noisy systems, the detection results using the fixed step size method still have residual errors compared to the optimal linear solution, requiring additional iterations or other methods to partially compensate for them. The EDRID method proposed in this invention effectively eliminates the convergence error bound by utilizing a dynamic step size mechanism: according to the derivation, the formula for the upper bound of the convergence error is... At that time, the expected error of the EDRID method after convergence is approximately By employing a dynamic step size, the aforementioned error terms are gradually canceled out, allowing the expected error to arbitrarily approach 0. In other words, the EDRID method theoretically achieves unbiased convergence to the least squares solution, and after sufficient iterations, it can achieve detection accuracy almost identical to that of directly solving the LS solution. Furthermore, the derivation formula for the upper bound of the convergence error proves that the EDRID method with a dynamic step size satisfies... The algorithm has a global convergence condition, so regardless of the initialization, the iterations are guaranteed to converge to the correct solution, improving the stability and robustness of the algorithm. In contrast, although the fixed step size method can also approach the LS solution by increasing the number of iterations, it lacks a similar adaptive convergence guarantee. Under different channel conditions, it may be necessary to manually adjust the step size or the number of iterations to balance convergence speed and accuracy.

[0181] In summary, the EDRID method proposed in this invention maintains the advantages of existing schemes while achieving lower computational complexity and higher convergence accuracy through pseudo-inverse operations and dynamic step-size correction. Theoretically, the EDRID method proposed in this invention proves convergence to the global LS solution and outperforms existing schemes such as the MCRBK algorithm in terms of complexity, error control, and stability. In other words, the method of this invention can achieve performance comparable to optimal linear detection with lower computational and communication costs, demonstrating significant innovation and practical value.

[0182] Example 2

[0183] This invention provides a low-complexity distributed intelligent cooperative MIMO signal processing system, comprising several distributed units and a central processing unit; each distributed unit has a bidirectional interactive link with the central processing unit; each distributed unit includes a receiving antenna array, a radio frequency processing unit, a channel estimation unit, and a signal detection unit;

[0184] Each distributed unit executes a preset detection step until the preset number of iterations reaches a preset threshold. ;;

[0185] The central processing unit receives the data. The estimated vector of the next iteration The estimated vector is then quantized to obtain the final detection signal. ;

[0186] The preset detection steps include:

[0187] The receiving antenna array is used to receive radio frequency signals;

[0188] The radio frequency processing unit processes the received radio frequency signal to obtain the received signal vector;

[0189] The channel estimation unit uses a known pilot sequence to perform local channel estimation and obtain a local channel matrix.

[0190] The signal detection unit calculates the dynamic step size according to a predetermined adaptive dynamic step size adjustment strategy; updates the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration; selects the next distributed unit using a multi-step conditional random sampling strategy; and sends the updated estimated vector to the selected next distributed unit.

[0191] In one specific embodiment of the present invention, the step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes:

[0192] In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows:

[0193] ,

[0194] In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

[0195] In one specific embodiment of the present invention, the formula for calculating the dynamic step size is:

[0196] ,

[0197] In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

[0198] In one specific embodiment of the present invention, the method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration is calculated using the following formula:

[0199] ,

[0200] In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

[0201] In one specific embodiment of the present invention, the final detection signal The calculation formula is:

[0202] ,

[0203] In the formula, Round to the nearest whole number constellation points, for A set consisting of constellation points corresponding to each transmitted signal.

[0204] The rest are the same as in Example 1.

[0205] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0206] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0207] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0208] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0209] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0210] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A low-complexity distributed intelligent collaborative MIMO signal processing method, characterized in that, include: The selected distributed unit performs a preset detection step to obtain the estimated vector until the number of iterations reaches a preset iteration threshold. Each distributed unit is accessed once within a sampling period; Will pass The estimated vector of the next iteration The data is sent to the central processing unit (CPU) for quantization and vector estimation. The final detection signal is obtained. ; The preset detection steps include: The received radio frequency signal is processed to obtain the received signal vector; Local channel estimation is performed using known pilot sequences to obtain the local channel matrix; The dynamic step size is calculated based on the predetermined adaptive dynamic step size adjustment strategy. The estimated vector is updated using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration. The next distributed unit is selected using a multi-step conditional random sampling strategy. The updated estimated vector is sent to the next selected distributed unit.

2. The low-complexity distributed intelligent cooperative MIMO signal processing method according to claim 1, characterized in that: The step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes: In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows: ， In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

3. The low-complexity distributed intelligent cooperative MIMO signal processing method according to claim 1, characterized in that: The formula for calculating the dynamic step size is: ， In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

4. The low-complexity distributed intelligent cooperative MIMO signal processing method according to claim 1, characterized in that: The method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration employs the following calculation formula: ， In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Represents the local channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

5. The low-complexity distributed intelligent cooperative MIMO signal processing method according to claim 1, characterized in that: The final detection signal The calculation formula is: ， In the formula, Round to the nearest whole number constellation points, for A set consisting of constellation points corresponding to each transmitted signal.

6. A low-complexity distributed intelligent cooperative MIMO signal processing system, characterized in that, It includes several distributed units and a central processing unit; each distributed unit has a bidirectional interactive link with the central processing unit; each distributed unit includes a receiving antenna array, a radio frequency processing unit, a channel estimation unit, and a signal detection unit; Each distributed unit executes a preset detection step until the preset number of iterations reaches a preset threshold. ;; The central processing unit receives the data. The estimated vector of the next iteration and for the estimated vector Quantization processing is performed to obtain the final detection signal. ; The preset detection steps include: The receiving antenna array is used to receive radio frequency signals; The radio frequency processing unit processes the received radio frequency signal to obtain the received signal vector; The channel estimation unit uses a known pilot sequence to perform local channel estimation and obtain a local channel matrix. The signal detection unit calculates the dynamic step size according to a predetermined adaptive dynamic step size adjustment strategy; updates the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration; selects the next distributed unit using a multi-step conditional random sampling strategy; and sends the updated estimated vector to the selected next distributed unit.

7. A low-complexity distributed intelligent cooperative MIMO signal processing system according to claim 6, characterized in that, The step of selecting the next distributed unit using a multi-step conditional random sampling strategy includes: In the In the next iteration, based on the preset exclusion... The selection of the next distributed unit based on the multi-step conditional sampling probability calculation formula is used to select the next distributed unit. The calculation formula is as follows: ， In the formula, Indicates the first In the next iteration, exclude the previous ones. The selection of the next distributed unit based on the multi-step conditional sampling probability of the sampled result. This represents the sampling probability of a general random block sampling method. Indicates the number of iterations. The sampling result, i.e., the index of the selected distributed unit, Indicates the first The index of the distributed unit selected in the next iteration. Indicates the first The distributed unit selected in the next iteration.

8. A low-complexity distributed intelligent cooperative MIMO signal processing system according to claim 6, characterized in that, The formula for calculating the dynamic step size is: ， In the formula, For the first The dynamic step size corresponding to the next iteration. This represents the total number of transmit antennas in a MIMO system. This represents the total number of receiving antennas in a MIMO system. For the number of iterations, This represents the total number of receiving antennas in a single distributed unit.

9. A low-complexity distributed intelligent cooperative MIMO signal processing system according to claim 6, characterized in that, The method for updating the estimated vector using the received signal vector, the local channel matrix, the dynamic step size, and the estimated vector obtained after the previous iteration employs the following calculation formula: ， In the formula, For the first The updated estimated vector corresponding to the next iteration. For the first The updated estimated vector corresponding to the next iteration. For the local channel matrix, Represents the local channel matrix The conjugate transpose of . For the received signal vector, For the first The dynamic step size corresponding to the next iteration. Indicates the first The index of the distributed unit selected in the next iteration.

10. A low-complexity distributed intelligent cooperative MIMO signal processing system according to claim 6, characterized in that, The final detection signal The calculation formula is: ， In the formula, Round to the nearest whole number constellation points, for A set consisting of constellation points corresponding to each transmitted signal.

Images